## wpiea2021275-print-pdf

## Source details

**Canonical URL:** [wpiea2021275-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021275-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021275-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021275-print-pdf.pdf.json)

---

### Introduction: role, evidence, and research question
- The IMF assists member countries in implementing macroeconomic programs that smooth external adjustment, typically disbursing financing tranches as policy conditionality is met.
- A realistic set of projections for macroeconomic variables such as output and inflation is an important backbone of a successful and credible program.
- Across 306 programs approved from 2000 to 2019, the mean annual growth rate over the next two years ended up being 0.76 percentage points lower than initially forecast at the time of the program request.
- The level of output two years hence ended up being 1.5 percentage points lower than forecast.
- Over-optimism is observed in median forecasts for programs implemented in both advanced and emerging economies (GRA-financed) and those made available on concessional terms to low-income members (PRGT-financed).
- Forecast over-optimism may reflect:
  - unforeseen shocks whose realizations in a sample do not have zero mean;
  - inefficiency if forecasters do not make full use of information available at the time of the forecast (e.g., planned policy changes do not have expected impacts, planned policy changes are not implemented, or forecasters mis-model relationships between past and future observables).
- Focus: test whether ex post growth forecast errors are correlated with financial system characteristics observable when a program was designed, with primary interest in the credit-to-GDP gap (deviation of ratio of credit to private sector to GDP from its trend).
- Observed pattern: for GRA programs in the top quartile of growth forecast errors, the median credit-to-GDP gap increases by over 6 percentage points in the three years prior to program approval, then reverts sharply in following years; this pattern is not observed for low-optimism programs or for PRGT programs.
- Main empirical finding: financial variables observable during program design—particularly recent evolution of the credit-to-GDP gap—are reliable and strong predictors of growth forecast errors.

### Event-study of IMF-supported programs: sample, measurement, and main estimates
- Sample counts:
  - 306 IMF-supported programs approved 2000–2019 (129 GRA; 177 PRGT).
  - Baseline regression sample: 218 programs (89 GRA and 129 PRGT) with credit series available for at least seven years prior to approval.
- Outcome and forecast error definitions:
  - Growth outcome: annualized growth rate of real output over the two years following program approval, denoted g_{i,t}, computed as the difference in levels of real output from t-1 to t+1 and expressed in percent.
  - Forecast error ĝ_{i,t} defined as the forecast made at program approval minus the outturn (outturns from the October 2021 WEO vintage). Positive ĝ_{i,t} denotes an optimistic growth forecast.
- Descriptive statistics of ĝ_{i,t} (full sample 2000–19):
  - Average = 0.8.
  - Median = 0.4.
  - Standard deviation = 3.9.
  - One quarter of programs have ĝ_{i,t} within 0.68 percentage points of zero.
  - The 32 programs approved in 2008 and in 2019 have a median ĝ_{i,t} of nearly 3.1.
  - Some PRGT programs show annualized forecast errors upwards of 10 to 20 percentage points.
  - 13 programs with ĝ_{i,t} more than two standard deviations from zero; eight of these requests took place within three years of organized violent conflict.
- Baseline regressor: change in the credit-to-GDP gap from t-3 to t-1; gap computed as deviation from trend with trend a one-sided five-year moving average; GDP denominator uses WEO vintage immediately preceding program approval.
- Baseline regression (Table 1, column 1):
  - Coefficient on credit-to-GDP gap (change t-3 to t-1) = 0.07*** (0.02). R2 = 0.13. Observations = 218. Mean dependent variable = 0.70.
  - Interpretation: every percentage point increase in the credit-to-GDP gap in the two years prior to approval is associated with annualized two-year growth forecasts more optimistic by 0.075 percentage points.
  - A one standard deviation increase in the credit-to-GDP gap (approximately 8.1 percentage points) is associated with an annualized forecast error approximately 0.60 percentage points more optimistic, implying the level of output is over-estimated by about 1.2 percentage points at a two-year horizon.
- Subsample and robustness highlights (Table 1):
  - Optimistic forecasts only (ĝ_{i,t} ≥ 0): credit gap = 0.14** (0.06); observations = 128; mean dep var = 2.57; R2 = 0.15. One standard deviation increase implies over-estimation of level of output by about 2.3 percentage points at two years.
  - Credit gap > 0 in year prior: credit gap = 0.10** (0.05); observations = 160; R2 = 0.17.
  - GRA programs: credit gap = 0.07*** (0.02); observations = 89; R2 = 0.22; mean dep var = 0.82.
  - PRGT programs: credit gap = 0.08 (0.10); observations = 129; R2 = 0.13; mean dep var = 0.63.
  - Post-GFC sample (2010–19): credit gap = 0.10** (0.05); observations = 105; R2 = 0.21; mean dep var = 1.32.
  - Including dummies for 2008 and 2011: credit gap = 0.06*** (0.02); observations = 218; R2 = 0.16.
- Additional findings:
  - Credit-to-GDP gap not statistically significant among programs with pessimistic forecasts (ĝ_{i,t} ≤ 0).
  - PRGT programs coinciding with violent conflict: annualized growth forecast error is 2.34** larger in relevant specification.
  - Robustness: three-year horizon yields similar results; excluding 2019 programs leaves results unchanged; excluding precautionary credit lines and non-financial programs leaves results unchanged.
- Table 2 (programmed policy adjustments included):
  - Selected estimates:
    - Column (1): credit gap = 0.15** (0.07); civil conflict = 2.05** (0.80); output growth (t-1) = 0.21*** (0.08); dummy 2019 = 3.38*** (1.11). Observations = 91. R2 = 0.27.
    - Column (3): credit gap = 0.08** (0.04); civil conflict = 1.80* (0.98); output growth (t-1) = 0.13 (0.09); dummy 2019 = 2.52*** (0.69). Observations = 136. R2 = 0.10.
  - Programmed policy variables (subsets): expected real growth of base money (t-1 to t+1) = -0.02 (0.04); expected change in money multiplier (t-1 to t+1) = 0.18 (0.88); expected fiscal adjustment (t-1 to t+1) = 0.17 (0.14) — not statistically significant.
- Core analytical conclusion from event study:
  - Information in the credit-to-GDP gap observable at program approval is not fully incorporated into IMF program growth forecasts, producing systematic optimistic forecast errors associated with prior credit expansions.
  - Economic magnitude reiterated: one standard deviation increase in credit gap (~8.1 percentage points) → annualized two-year forecast error ~0.60 percentage points more optimistic → over-estimation of level of output ~1.2 percentage points at two years.
  - Inefficiency concentrated among optimistic programs and those preceded by positive credit gaps; present in both GRA and PRGT in magnitude but clearer statistically for GRA.
  - Relationship persists in post-Global Financial Crisis period, with coefficient increasing to 0.10 in 2010–19.

### Exploring channels of macrofinancial inefficiency: programmed adjustment, implementation, and interpretation
- Hypothesis tested: pre-program credit expansions lead to restrictive program policy conditionality, and forecast optimism arises because such restrictive policies are politically challenging to implement or their growth impact is underestimated.
- Programmed monetary adjustment proxy (expected growth of monetary base deflated by expected CPI growth): not significantly correlated with growth forecast errors; credit-gap effect unchanged.
- Programmed financial-conditions proxy (expected change in money multiplier): not significantly correlated with growth forecast errors.
- Programmed fiscal adjustment proxy (expected change in overall fiscal balance t-1 to t+1, truncated to [-10,10] percentage points): planned fiscal adjustments appear positively correlated with forecast optimism, but not statistically significant.
- Conclusion: programmed adjustments do not explain inefficiency; endogenous calibration of program conditionality in response to credit conditions is unlikely to be the primary source.
- Tests of program implementation (off-track programs):
  - Forecast errors for monetary and financial-adjustment proxies are not significantly correlated with pre-program credit developments.
  - Policy-adjustment forecast errors not significantly correlated with degree of planned adjustment.
  - Conclusion: implementation failure (programs going “off track”) is unlikely to explain inefficiency.
- Alternative explanation favored:
  - Forecasters have difficulty distinguishing in real time between structural financial development and cyclical credit booms and are biased toward structural interpretation, underestimating mean reversion of private credit and its impact on growth.
- Panel analysis (WEO and MONA forecasts; Table 3 highlights):
  - Program forecasts (column 1; 1,425 obs; 2003–19): credit gap = 0.06*** (0.02); FE trade partners’ growth (t+1) = 0.50*** (0.06); real GDP growth (t-1) = 0.09** (0.04); R2 = 0.23; mean dep var = 0.00.
  - No-program forecasts (column 2; 2,374 obs): credit gap = 0.02** (0.01); FE trade partners’ growth = 0.50*** (0.05); R2 = 0.26; mean dep var = 0.32.
  - All forecasts (column 3; 3,799 obs): credit gap = 0.03*** (0.01); FE trade partners’ growth = 0.50*** (0.04); R2 = 0.24; mean dep var = 0.20.
  - Top-quartile optimistic forecasts (column 4; 948 obs; ĝ ≥ 1.3): credit gap = 0.03*** (0.01); share of loans denom. FX (t-1) = 0.01** (0.01) — among most optimistic forecasts FX loan share matters.
  - Consensus forecasts (column 5; 2,056 obs): results consistent with baseline — inefficiency present in private-sector forecasts.
  - REER gap and interaction with high FX loans (column 6): REER gap = 0.05*** (0.01); interaction REER * Hi FX loans = -0.05** (0.02); credit gap significance unchanged.
  - Banking-crisis dummies (column 7): banking crisis in t or t+1 = 1.21*** (0.41) — crises during forecast horizon strongly associated with optimistic forecast errors; pre-forecast banking crises not significant; credit gap remains significant after controlling for crises.
- Overall panel interpretation:
  - Recent evolution of the credit-to-GDP gap is a robust predictor of forecast optimism across program and surveillance forecasts and private forecasts.
  - Forecast errors of trading partners’ growth are tightly correlated with domestic forecast errors (FE trade partners’ growth coefficient ≈ 0.50***).
  - Liability dollarization modifies the relationship between REER gaps and forecast errors.

### Robustness, fixed effects, and synthesis of policy implications
- Robustness to alternative fixed effects (Table 4 reproducing baseline):
  - Credit-to-GDP gap (change t-3 to t-1) estimate robust across specifications: 0.03*** (0.01) in columns with None, Country, Year, Country & Year, and Country & Vintage fixed effects.
  - Observations = 3,799 in all specifications; time period = 2003–19.
  - R-squared ranges: 0.24; 0.38; 0.30; 0.42; 0.55 (columns 1–5).
  - Other controls show varied significance; FE trade partners’ growth generally highly significant.
- Conclusions and policy implications:
  - This paper is the first to study the role of financial variables in explaining optimism in growth forecasts made in IMF-supported programs.
  - Commonly available financial variables are correlated with subsequent forecast errors made in program design.
  - An expansion of the credit-to-GDP gap in the years preceding a program request is associated with a significantly more optimistic growth forecast for the next two years.
  - The result is strongest among forecasts that were most optimistic; errors are also increasing in the economy’s degree of liability dollarization.
  - The inefficiency does not reflect endogenous calibration of monetary and financial policies in response to credit conditions (this channel is tested and rejected).
  - Similar inefficiencies are documented in the IMF’s surveillance forecasts and in private professional forecasts.
  - Suggested mechanism: difficulty in real time distinguishing structural versus cyclical credit expansions; further research needed into distinguishing credit dynamics in real time and developing models of their short-term macroeconomic implications to better inform growth forecasts.

*Source: IMF working paper — content unit wpiea2021275-print-pdf.*

### 1.          I

### 1. INTRODUCTION

### Role of the IMF and importance of forecast realism
- The IMF assists member countries in implementing macroeconomic programs that smooth external adjustment, typically disbursing financing tranches as policy conditionality is met.
- A realistic set of projections for macroeconomic variables such as output and inflation is an important backbone of a successful and credible program.

### Evidence of over-optimism in IMF program forecasts
- Across 306 programs approved from 2000 to 2019, the mean annual growth rate over the next two years ended up being 0.76 percentage points lower than initially forecast at the time of the program request.
- The level of output two years hence ended up being 1.5 percentage points lower than forecast.
- Over-optimism is observed in median forecasts for programs implemented in both advanced and emerging economies (GRA-financed) and those made available on concessional terms to low-income members (PRGT-financed).
- Figure notes: medians across 129 IMF-supported programs supported by GRA resources and 177 programs supported by PRGT resources approved between 2000 and 2019.

### Potential sources of forecast bias and inefficiency
- Forecast over-optimism may reflect:
  - Unforeseen shocks (e.g., fall in world commodity prices, civil conflict, global financial crisis) whose realizations in a sample do not have zero mean.
  - Inefficiency: forecasters not making full use of information available at the time of the forecast. This can arise if:
    - Planned policy changes do not have expected impacts on growth.
    - Planned policy changes are not implemented.
    - Forecasters do not correctly model relationships between past and future observable variables.
- A biased forecast is considered inefficient if past bias predicts future errors.

### Focus on financial variables and the credit cycle
- The paper applies forecast efficiency tests that check whether ex post growth forecast errors are correlated with financial system characteristics observable when a program was designed.
- The main financial variable of interest is the credit-to-GDP gap (deviation of ratio of credit to private sector to GDP from its trend).
- Observed pattern: for GRA programs that were most optimistic (top quartile of growth forecast errors), the median credit-to-GDP gap increases by over 6 percentage points in the three years prior to program approval, then reverts sharply in following years. This pattern is not observed for low-optimism programs or for PRGT programs.

### Key empirical contribution and results
- Methodology builds on Blanchard and Leigh (2013) and related literature examining inefficiency in program forecasts.
- Main result: financial variables observable during program design—particularly recent evolution of the credit-to-GDP gap—are reliable and strong predictors of growth forecast errors.
- Quantitative findings:
  - A program preceded by a one standard deviation increase in the credit-to-GDP gap—approximately 8.1 percentage points in the sample—is associated with an annualized forecast error that is approximately 0.60 percentage points more optimistic.
  - This implies the level of output is over-estimated by about 1.2 percentage points at a two-year horizon.
  - The effect is about twice as strong among programs whose growth forecasts turned out to be optimistic.

### Exploration of mechanisms and interpretation
- Considered plausible mechanisms for inefficient use of financial information:
  - Endogenous policy adjustments in programs following credit booms calibrated to tighten financial conditions, with potential underestimation of their growth impact.
  - Political economy: restrictive policies may be less likely to be implemented, making programs after credit booms more likely to go “off track.”
- Evidence presented suggests these mechanisms are unlikely to fully explain the result.
- More likely interpretation: forecasters do not correctly model the relationship between past and future evolution of observable variables—possibly underestimating credit’s tendency to revert to trend—and the interactions between the credit cycle and the real economy.

### Broader applicability and institutional implications
- The result also applies outside IMF-supported programs:
  - Detected in a large panel of WEO forecasts in the IMF’s surveillance role and in forecasts by professional forecasters.
  - The size of the effect appears smaller than in the event study of IMF-supported programs.
- Institutional implication: findings underscore the importance of continuing to deepen macrofinancial analysis (as highlighted in the Comprehensive Surveillance Review) and incorporating such analysis into program design to reduce optimism in growth forecasts.

### Organization of the paper
- Section 2: documents stylized facts about growth forecast errors in IMF-supported programs and presents the event-study empirical approach.
- Section 3: presents a second empirical approach covering a broad panel of IMF forecasts (program and surveillance contexts) and forecasts by professional forecasters.
- Section 4: concludes.

*Source: IMF working paper — Introduction (content unit: wpiea2021275-print-pdf - 1.          I).*

### 2.          E

### 2. E VENT STUDY OF IMF-SUPPORTED PROGRAMS

### Sample and measurement
- Between 2000 and 2019 there were 306 IMF-supported programs approved, of which 129 were supported by GRA resources and 177 by concessional PRGT resources.
- Baseline estimation sample for regressions: 218 programs (89 GRA and 129 PRGT) for which credit aggregate series are available for at least seven years prior to program approval.
- Growth outcome measure: annualized growth rate of real output over the two years following program approval, denoted 푔
௖,௜, computed as the difference in levels of real output from t-1 to t+1 and expressed in percent.
- Forecast error 푔ො
௜
 defined as the forecast made at program approval minus the outturn (outturns from the October 2021 WEO vintage). Positive 푔ො
௜
 denotes an optimistic growth forecast (growth turned out to be lower than anticipated); negative denotes pessimistic.

### Descriptive statistics and salient patterns
- Average of 푔ො
௜
 in the full sample (2000–19): 0.8.
- Median of 푔ො
௜
 in the full sample: 0.4.
- Standard deviation of 푔ො
௜
 in the full sample: 3.9.
- One quarter of programs have 푔ො
௜
 within 0.68 percentage points of zero.
- The within-year median 푔ො
௜
 is generally close to zero except in years associated with major global crises (notably, 2008 and 2019).
- The 32 IMF-supported programs approved in 2008 and in 2019 have a median 푔ො
௜
 of nearly 3.1.
- Programs in low-income economies supported by PRGT exhibit larger variance of forecast errors, with some annualized forecast errors upwards of 10 to 20 percentage points.
- There are 13 programs with 푔ො
௜
 more than two standard deviations from zero; eight of these program requests took place within three years of an episode of organized violent conflict (UCDP data).

### Baseline empirical specification and identification
- Event-study regression: regress program i’s two-year annualized growth forecast error (푔ො
௜
) on observable information at approval.
- Key regressor (in baseline 푿
௧ିଵ,௜
): change in the credit-to-GDP gap over the two years preceding program approval (from t-3 to t-1). The credit-to-GDP gap is computed as the deviation of credit-to-GDP from its trend; trend is a one-sided five-year moving average. GDP denominator uses WEO vintage immediately preceding program approval to preserve real-time information set.
- Controls/dummies included: dummy for violent conflict episode within three years of program approval; lagged output growth (t-1); dummy for programs approved in the year preceding Covid-19 (2019); additional robustness dummies for 2008 and 2011 in some specifications.

### Main regression results (summary of Table 1)
- Baseline coefficient (full sample, column 1): Credit-to-GDP gap (change from t-3 to t-1) = 0.07*** (standard error 0.02). R2 = 0.13. Observations = 218. Mean of dependent variable (푔ො
௜
) = 0.70.
- Interpretation: For every percentage point that the credit-to-GDP gap increases in the two years prior to program approval, annualized two-year growth forecasts are more optimistic by 0.075 percentage points (reported text value). A one standard deviation increase in the credit-to-GDP gap—approximately 8.1 percentage points in the sample—is associated with an annualized forecast error approximately 0.60 percentage points more optimistic, implying the level of output is over-estimated by about 1.2 percentage points at a two-year horizon.
- Column 2 (sample with optimistic forecasts only, 푔ො
௜
≥0): Credit-to-GDP gap = 0.14** (0.06). Observations = 128. Mean of dependent variable = 2.57. R2 = 0.15. Text interpretation: coefficient nearly twice baseline; one standard deviation increase associated with over-estimation of level of output by about 2.3 percentage points at two-year horizon.
- Column 3 (sample with credit gap > 0 in year prior): Credit-to-GDP gap = 0.10** (0.05). Observations = 160. R2 = 0.17. Mean dep var = 0.68.
- Columns separating funding type:
  - GRA programs (column 4): Credit-to-GDP gap = 0.07*** (0.02). Observations = 89. R2 = 0.22. Mean dep var = 0.82.
  - PRGT programs (column 5): Credit-to-GDP gap = 0.08 (0.10). Observations = 129. R2 = 0.13. Mean dep var = 0.63.
  - Text interpretation: similar magnitudes across groups but statistical significance confirmed primarily in GRA group.
- Column 6 (includes dummies for 2008 and 2011): Credit-to-GDP gap = 0.06*** (0.02). Observations = 218. R2 = 0.16. Dummy 2008 and 2011 capture significant common forecast optimism but coefficient on gap remains significant.
- Column 7 (post-GFC sample, 2010–19): Credit-to-GDP gap = 0.10** (0.05). Observations = 105. R2 = 0.21. Mean dep var = 1.32. Text interpretation: coefficient increases to 0.10 in post-GFC period, indicating forecasters have not become more efficient at incorporating financial information.

### Additional findings and robustness
- The credit-to-GDP gap is not statistically significant among programs that ended up with pessimistic forecasts (푔ො
௜
≤0).
- Among the 58 programs preceded by a negative credit-to-GDP gap (results not reported), the coefficient is positive but not statistically significant—inefficiency strongest when preceded by a positive credit gap.
- PRGT-supported programs that coincide with violent conflict: annualized growth forecast error is 2.34** (column 5) larger; conflict variable not significant for GRA-supported programs.
- Robustness checks:
  - Three-year horizon for forecast error yields similar results, significant at conventional levels though somewhat weaker economic magnitude.
  - Excluding programs approved in 2019 (Covid-19 affected outturns) leaves results unchanged.
  - Excluding precautionary credit lines and non-financial programs leaves results unchanged.

### Results including effect of programmed policy adjustments (Table 2)
- Table 2 column summaries (selected):
  - Column (1): Credit-to-GDP gap = 0.15** (0.07). Dummy: Civil conflict = 2.05** (0.80). Output growth (t-1) = 0.21*** (0.08). Dummy: 2019 = 3.38*** (1.11). Observations = 91. R2 = 0.27. Time period 2002-19.
  - Column (2): Credit-to-GDP gap = 0.16** (0.08). Dummy: Civil conflict = 2.02** (0.81). Output growth (t-1) = 0.21*** (0.08). Dummy: 2019 = 3.49*** (0.96). Observations = 92. R2 = 0.27. Time period 2002-19.
  - Column (3): Credit-to-GDP gap = 0.08** (0.04). Dummy: Civil conflict = 1.80* (0.98). Output growth (t-1) = 0.13 (0.09). Dummy: 2019 = 2.52*** (0.69). Observations = 136. R2 = 0.10. Time period 2000-19.
- Additional programmed policy variables tested (in subsets): expected real growth of base money (t-1 to t+1) = -0.02 (0.04); expected change in money multiplier (t-1 to t+1) = 0.18 (0.88); expected fiscal adjustment (t-1 to t+1) = 0.17 (0.14). These variables are not reported as statistically significant in the table notes.

### Core analytical conclusions
- Growth forecast inefficiency: Information in the credit-to-GDP gap observable at program approval is not fully incorporated into IMF program growth forecasts, producing systematic optimistic forecast errors associated with prior credit expansions.
- Economic significance: A one standard deviation increase in the credit-to-GDP gap (about 8.1 percentage points) implies an annualized two-year forecast error about 0.60 percentage points more optimistic and an over-estimation of output level by about 1.2 percentage points at two years.
- The inefficiency is concentrated among programs that turn out to be optimistic and among programs preceded by positive credit gaps; it is present in both GRA and PRGT samples in magnitude but clearer statistically for GRA programs.
- Episodes of violent conflict and major global crises (2008, 2019) are associated with substantially larger forecast errors; in PRGT programs, conflict is strongly associated with greater optimism.
- The relationship between past credit expansions and forecast optimism persists in the post-Global Financial Crisis period (2010–19), with the coefficient on the credit gap increasing to 0.10, suggesting forecasters have not materially improved incorporation of financial variables into growth forecasts.

*Source: Authors’ calculations using MONA and October 2021 vintage of the World Economic Outlook; event-study regression results reported in Tables 1 and 2 of the chapter.*

### 3.          E

### 3.          E
XPLORING POTENTIAL CHANNELS OF MACROFINANCIAL INEFFICIENCY

### Hypothesis and approach
- Hypothesis: pre-program credit expansions are met with particularly restrictive program policy conditionality, and these policy adjustments may lead to forecast optimism if they are politically challenging to implement or if their impact on growth is underestimated.
- Empirical approach: augment baseline growth-forecast-error equation to include measures of expected monetary, financial, and fiscal policy adjustment (results reported in Table 2 and Table 3).

### Tests of programmed policy adjustment as an explanation
- Programmed monetary adjustment proxy:
  - Expected rate of growth of the monetary base over the next two years, deflated by expected CPI growth.
  - Finding: this variable is not significantly correlated with growth forecast errors; the economic and statistical significance of the pre-program credit-to-GDP gap is unaffected.
- Programmed financial-conditions adjustment proxy:
  - Expected change in the money multiplier (ratio of broad money to the monetary base) from t-1 to t+1.
  - Finding: this variable is not significantly correlated with growth forecast errors.
- Programmed fiscal adjustment proxy:
  - Expected change in the overall fiscal balance as a share of GDP from t-1 to t+1 (truncated to lie between -10 and 10 percentage points of GDP; two programs excluded).
  - Finding: planned fiscal adjustments appear positively correlated with forecast optimism (suggesting fiscal multipliers may be somewhat underestimated), but this result is not statistically significant in this sample.
- Conclusion from programmed-adjustment tests:
  - Even after controlling for programmed adjustment of monetary, financial, and fiscal policies, forecast optimism remains significantly correlated with the pre-program evolution of the credit-to-GDP gap.
  - Therefore, inefficiency is unlikely to be generated by endogenous calibration of program conditionality in response to pre-program credit conditions.

### Tests of program implementation (off-track programs) as an explanation
- Method: compute forecast errors for the two proxies of monetary and financial policy adjustment and regress these on pre-program credit developments; test whether policy-adjustment forecast errors correlate with degree of planned adjustment.
- Finding: neither monetary-adjustment nor financial-adjustment forecast errors are significantly correlated with pre-program credit developments; policy-adjustment forecast errors are not significantly correlated with the degree of planned policy adjustment.
- Conclusion: inefficiency is unlikely to be explained by the degree of program implementation (i.e., programs going “off track”).

### Alternative explanation: structural vs cyclical interpretation of credit expansion
- Hypothesis: forecasters have difficulty distinguishing in real time between structural financial development and cyclical credit booms.
- Interpretation: forecasters are biased toward the structural interpretation of credit expansions, failing to forecast either the mean reversion of private credit or the impact of cyclical credit reversion on growth.

### Panel analysis of growth forecasts (section 4) — setup
- Data:
  - Semi-annual WEO vintages for 130 countries and 497 MONA program forecasts over sample period.
  - Unit: country c, forecast vintage v, year t; dependent variable 푔ො
௖,௧ାଵ
௩ (forecast error for annualized growth rate in years t and t+1).
  - Sample means/medians of 푔ො
௖,௧ାଵ
௩: program sample mean 0.00 and median -0.11; no-program sample mean 0.32 and median 0.03.
- Baseline regressors (vector 푿
௖,௧ିଵ
풗):
  - Recent evolution of the credit-to-GDP gap (same definition as section 2; change from t-3 to t-1).
  - Lagged share of bank loans to non-financial corporates denominated in foreign currency (state variable).
- Vintage-specific controls (풀
௖,௧ାଵ
௩):
  - Ex post forecast error for trading partners’ growth (from GEE).
  - Expected change in fiscal balance as share of GDP from t-1 to t+1 (vintage v).
  - Dummy for civil conflict in forecast period.
  - Growth of real output prior to program approval.
  - Ex post revisions to the level of real GDP in period t-1.

### Key panel results (Table 3 summaries)
- Column 1 (Program forecasts; 1,425 observations; 2003-19):
  - Credit-to-GDP gap (change from t-3 to t-1): coefficient 0.06*** (standard error (0.02)).
  - Share of loans denom. FX (t-1): coefficient -0.00 (0.00).
  - FE trade partners’ growth (t+1): coefficient 0.50*** (0.06).
  - Real GDP growth (t-1): coefficient 0.09** (0.04).
  - R2 = 0.23; Mean of dependent variable 푔ො
௖,௧
௩ = 0.00.
  - Interpretation: recent evolution of the credit-to-GDP gap is significantly correlated with forecast errors; a 1-unit change in the gap is associated with a 0.06 change in annualized forecast error in this sample.
- Column 2 (No-program forecasts; 2,374 observations):
  - Credit-to-GDP gap: coefficient 0.02** (0.01).
  - FE trade partners’ growth: coefficient 0.50*** (0.05).
  - R2 = 0.26; Mean of dependent variable 푔ො
௖,௧
௩ = 0.32.
  - Interpretation: effect remains in surveillance forecasts but is smaller (coefficient 0.02).
- Column 3 (All forecasts; 3,799 observations):
  - Credit-to-GDP gap: coefficient 0.03*** (0.01).
  - FE trade partners’ growth: coefficient 0.50*** (0.04).
  - R2 = 0.24; Mean of dependent variable 푔ො
௖,௧
௩ = 0.20.
  - Interpretation: inefficiency documented in Fund-supported programs holds across all WEO and MONA forecasts.
- Column 4 (Top-quartile optimistic forecasts; 948 observations; 푔ො
௖,௧
௩ ≥ 1.3):
  - Credit-to-GDP gap: coefficient 0.03*** (0.01).
  - Share of loans denom. FX becomes positively correlated with forecast optimism: coefficient 0.01** (0.01).
  - Interpretation: among most optimistic forecasts, FX loan share matters; a one standard-deviation increase in FX share (≈27 percentage points here) associates with a 0.3 percentage point more optimistic annualized growth forecast error.
- Column 5 (Consensus Forecasts sample; 2,056 observations):
  - Results consistent with baseline: inefficient use of financial information is not confined to IMF forecasts but also present in private-sector forecasts.
- Column 6 (REER gap and interaction with high FX loans):
  - REER gap (change from t-3 to t-1): coefficient 0.05*** (0.01).
  - Interaction: REER * Hi FX loans: coefficient -0.05** (0.02).
  - Interpretation: REER gap is positively correlated with forecast optimism when liability dollarization is low; when liability dollarization is high, the relationship is negated.
  - Credit-to-GDP gap significance remains unchanged.
- Column 7 (Banking-crisis dummies):
  - Dummy: banking crisis in two years prior to forecast: coefficient 0.06 (0.51) — not significant.
  - Dummy: banking crisis in t or t+1: coefficient 1.21*** (0.41) — significant (crises during forecast horizon strongly associated with optimistic forecast errors).
  - Credit-to-GDP gap remains significantly correlated with forecast errors even after controlling for banking crises.
- Other notable coefficients and findings:
  - Ex post revision to real GDP (t-1) has mixed significance across columns; e.g., 0.21** (0.11) in column 2.
  - Expected fiscal adjustment (t-1 to t+2) is generally not significantly associated with forecast errors in panel estimates.
  - Civil conflict dummy generally not significant.
- Overall interpretation:
  - The recent evolution of the credit-to-GDP gap is a robust predictor of forecast optimism across program and surveillance forecasts, private forecasts, and even after controlling for exchange rate misalignment, FX liability, and banking crises.
  - Forecast errors of trading partners’ growth are tightly correlated with domestic forecast errors (FE trade partners’ growth coefficient ≈ 0.50***).

### Mechanisms and interpretations
- Programmed policy adjustments (monetary, financial, fiscal) do not explain the identified inefficiency.
- Implementation failure of programmed adjustments is not supported as the primary mechanism.
- Forecasters likely overweight a structural interpretation of credit expansions and under-appreciate cyclical mean reversion and its growth impact.
- Exchange rate misalignment matters for forecast optimism when liability dollarization is low; liability dollarization modifies how REER gaps relate to forecast errors.
- Banking crises occurring during the forecast horizon sharply increase optimistic forecast errors, but the predictive power of pre-program credit developments remains outside crisis-triggering situations.

*Italic: Source — content unit "3.          E" from wpiea2021275-print-pdf (IMF).*

### 5.          R

### 5.          R
OBUSTNESS

### Robustness tests with alternative fixed effects
- Baseline specification reproduced from Table 3, column 3; estimate of equation 3 in the full sample of IMF forecasts.
- Alternative fixed effects introduced across specifications:
  - Column (1): None
  - Column (2): Country
  - Column (3): Year
  - Column (4): Country & Year
  - Column (5): Country & Vintage
- Key estimated coefficients and standard errors (as reported in Table 4):
  - Credit-to-GDP gap (change from t-3 to t-1): 0.03***  (0.01)
  - Share of loans denom. FX (t-1): -0.00  (0.00) in (1); 0.00  (0.01) in (2); -0.00  (0.00) in (3); 0.00  (0.01) in (4); -0.00  (0.01) in (5)
  - Dummy: Civil conflict: -0.14  (0.25); 0.29  (0.30); -0.11  (0.25); 0.28  (0.28); 0.25  (0.32)
  - Exp. fiscal adjustment (t-1 to t+2): -0.01  (0.03); -0.02  (0.03); 0.02  (0.03); 0.02  (0.03); 0.00  (0.03)
  - FE trade partners’ growth (t+1): 0.50***  (0.04); 0.54***  (0.04); 0.09  (0.08); 0.17**  (0.08); 0.19**  (0.09)
  - Ex post revision to real GDP (t-1): 0.08  (0.07); -0.04  (0.07); 0.07  (0.07); -0.06  (0.07); 0.06  (0.09)
  - Real GDP growth (t-1): 0.08**  (0.03); 0.15***  (0.03); 0.05*  (0.03); 0.16***  (0.03); 0.16***  (0.04)
- Constants by specification:
  - (1): -0.05  (0.14)
  - (2): -0.09  (0.11)
  - (3): -1.68***  (0.39)
  - (4): -1.50***  (0.35)
  - (5): -0.96**  (0.44)
- Sample and fit statistics:
  - Observations: 3,799 (in all columns)
  - R-squared: 0.24; 0.38; 0.30; 0.42; 0.55 (columns 1–5, respectively)
  - Time period: 2003-19
- Notes from table:
  - Dependent variable: 푔ො
௖,௧ାଵ
௩
, the forecast error for annualized growth of real output from t-1 to t+1.
  - White heteroskedasticity-robust standard errors clustered by country are shown in parentheses.
  - Significance notation: *** 0.01; ** 0.05; * 0.10.
  - Sample includes all forecast vintages from WEO and MONA databases, covering 130 countries.
- Robustness conclusion: Introducing country, year, country & year, or country & forecast vintage fixed effects does not affect the point estimate and standard error of the lagged credit-to-GDP gap variable (0.03***  (0.01) across specifications).

### Conclusions and policy implications
- Contribution:
  - This paper is the first to study the role of financial variables in explaining optimism in growth forecasts made in IMF-supported programs.
- Main findings:
  - Commonly available financial variables are correlated with subsequent forecast errors made in the design of IMF-supported programs.
  - An expansion of the credit-to-GDP gap in the years preceding a program request is associated with a significantly more optimistic growth forecast for the next two years.
  - The result is strongest among forecasts that were most optimistic; errors are also increasing in the economy’s degree of liability dollarization.
  - The inefficiency does not reflect endogenous calibration of monetary and financial policies in response to credit conditions (the authors test and reject this channel).
  - Similar inefficiencies are documented in growth forecasts produced in the IMF’s surveillance activities and in forecasts made by professional forecasters in the private sector.
- Suggested mechanisms and research needs:
  - A likely mechanism is the difficulty in real time of distinguishing whether a credit expansion is structural (financial development) or a cyclical boom.
  - Additional research is needed into distinguishing credit dynamics in real time and developing models of their short-term macroeconomic implications to better inform growth forecasts.

*wpiea2021275-print-pdf - 5.          R*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021275-print-pdf.pdf_